[SCSI] lpfc 8.3.33: Parallelize SLI-4 Q distribution

Commonize SLI-3/4 Ring/Queue framework, to keep SLI-3 compatibility
Parallelize SLI-4 Q distribution - to use multiple posting/completion queues

Signed-off-by: James Smart <james.smart@emulex.com>
Signed-off-by: James Bottomley <JBottomley@Parallels.com>

6 files changed, 77 insertions, 36 deletions

diff --git a/drivers/scsi/lpfc/lpfc.h b/drivers/scsi/lpfc/lpfc.h
index a65c05a8d48..a870af1b547 100644
--- a/drivers/scsi/lpfc/lpfc.h
+++ b/drivers/scsi/lpfc/lpfc.h
@@ -732,7 +732,7 @@ struct lpfc_hba {
 	uint32_t hbq_count;	        /* Count of configured HBQs */
 	struct hbq_s hbqs[LPFC_MAX_HBQS]; /* local copy of hbq indicies  */
 
-	uint32_t fcp_qidx;		/* next work queue to post work to */
+	atomic_t fcp_qidx;		/* next work queue to post work to */
 
 	unsigned long pci_bar0_map;     /* Physical address for PCI BAR0 */
 	unsigned long pci_bar1_map;     /* Physical address for PCI BAR1 */
diff --git a/drivers/scsi/lpfc/lpfc_hw.h b/drivers/scsi/lpfc/lpfc_hw.h
index 4ab0e3542c5..d74470481ca 100644
--- a/drivers/scsi/lpfc/lpfc_hw.h
+++ b/drivers/scsi/lpfc/lpfc_hw.h
@@ -1188,8 +1188,8 @@ typedef struct {
  */
 
 /* Number of rings currently used and available. */
-#define MAX_CONFIGURED_RINGS     3
-#define MAX_RINGS                4
+#define MAX_SLI3_CONFIGURED_RINGS     3
+#define MAX_SLI3_RINGS                4
 
 /* IOCB / Mailbox is owned by FireFly */
 #define OWN_CHIP        1
@@ -2993,7 +2993,7 @@ typedef struct _PCB {
 
 	uint32_t pgpAddrLow;
 	uint32_t pgpAddrHigh;
-	SLI2_RDSC rdsc[MAX_RINGS];
+	SLI2_RDSC rdsc[MAX_SLI3_RINGS];
 } PCB_t;
 
 /* NEW_FEATURE */
@@ -3103,18 +3103,18 @@ struct lpfc_pgp {
 
 struct sli2_desc {
 	uint32_t unused1[16];
-	struct lpfc_hgp host[MAX_RINGS];
-	struct lpfc_pgp port[MAX_RINGS];
+	struct lpfc_hgp host[MAX_SLI3_RINGS];
+	struct lpfc_pgp port[MAX_SLI3_RINGS];
 };
 
 struct sli3_desc {
-	struct lpfc_hgp host[MAX_RINGS];
+	struct lpfc_hgp host[MAX_SLI3_RINGS];
 	uint32_t reserved[8];
 	uint32_t hbq_put[16];
 };
 
 struct sli3_pgp {
-	struct lpfc_pgp port[MAX_RINGS];
+	struct lpfc_pgp port[MAX_SLI3_RINGS];
 	uint32_t hbq_get[16];
 };
 
diff --git a/drivers/scsi/lpfc/lpfc_init.c b/drivers/scsi/lpfc/lpfc_init.c
index 9efe5f82a6a..176302f0e02 100644
--- a/drivers/scsi/lpfc/lpfc_init.c
+++ b/drivers/scsi/lpfc/lpfc_init.c
@@ -4551,6 +4551,13 @@ lpfc_sli_driver_resource_setup(struct lpfc_hba *phba)
 			phba->cfg_sg_seg_cnt = LPFC_DEFAULT_MENLO_SG_SEG_CNT;
 	}
 
+	if (!phba->sli.ring)
+		phba->sli.ring = (struct lpfc_sli_ring *)
+			kzalloc(LPFC_SLI3_MAX_RING *
+			sizeof(struct lpfc_sli_ring), GFP_KERNEL);
+	if (!phba->sli.ring)
+		return -ENOMEM;
+
 	/*
 	 * Since the sg_tablesize is module parameter, the sg_dma_buf_size
 	 * used to create the sg_dma_buf_pool must be dynamically calculated.
@@ -4710,6 +4717,16 @@ lpfc_sli4_driver_resource_setup(struct lpfc_hba *phba)
 		sges_per_segment = 2;
 
 	/*
+	 * For SLI4, instead of using ring 0 (LPFC_FCP_RING) for FCP commands
+	 * we will associate a new ring, for each FCP fastpath EQ/CQ/WQ tuple.
+	 */
+	if (!phba->sli.ring)
+		phba->sli.ring = kzalloc(
+			(LPFC_SLI3_MAX_RING + phba->cfg_fcp_eq_count) *
+			sizeof(struct lpfc_sli_ring), GFP_KERNEL);
+	if (!phba->sli.ring)
+		return -ENOMEM;
+	/*
 	 * Since the sg_tablesize is module parameter, the sg_dma_buf_size
 	 * used to create the sg_dma_buf_pool must be dynamically calculated.
 	 * 2 segments are added since the IOCB needs a command and response bde.
@@ -5555,6 +5572,10 @@ lpfc_hba_free(struct lpfc_hba *phba)
 	/* Release the driver assigned board number */
 	idr_remove(&lpfc_hba_index, phba->brd_no);
 
+	/* Free memory allocated with sli rings */
+	kfree(phba->sli.ring);
+	phba->sli.ring = NULL;
+
 	kfree(phba);
 	return;
 }
@@ -6924,6 +6945,8 @@ lpfc_sli4_queue_destroy(struct lpfc_hba *phba)
 int
 lpfc_sli4_queue_setup(struct lpfc_hba *phba)
 {
+	struct lpfc_sli *psli = &phba->sli;
+	struct lpfc_sli_ring *pring;
 	int rc = -ENOMEM;
 	int fcp_eqidx, fcp_cqidx, fcp_wqidx;
 	int fcp_cq_index = 0;
@@ -7107,6 +7130,12 @@ lpfc_sli4_queue_setup(struct lpfc_hba *phba)
 				"rc = 0x%x\n", rc);
 		goto out_destroy_mbx_wq;
 	}
+
+	/* Bind this WQ to the ELS ring */
+	pring = &psli->ring[LPFC_ELS_RING];
+	pring->sli.sli4.wqp = (void *)phba->sli4_hba.els_wq;
+	phba->sli4_hba.els_cq->pring = pring;
+
 	lpfc_printf_log(phba, KERN_INFO, LOG_INIT,
 			"2590 ELS WQ setup: wq-id=%d, parent cq-id=%d\n",
 			phba->sli4_hba.els_wq->queue_id,
@@ -7137,6 +7166,12 @@ lpfc_sli4_queue_setup(struct lpfc_hba *phba)
 					"WQ (%d), rc = 0x%x\n", fcp_wqidx, rc);
 			goto out_destroy_fcp_wq;
 		}
+
+		/* Bind this WQ to the next FCP ring */
+		pring = &psli->ring[MAX_SLI3_CONFIGURED_RINGS + fcp_wqidx];
+		pring->sli.sli4.wqp = (void *)phba->sli4_hba.fcp_wq[fcp_wqidx];
+		phba->sli4_hba.fcp_cq[fcp_cq_index]->pring = pring;
+
 		lpfc_printf_log(phba, KERN_INFO, LOG_INIT,
 				"2591 FCP WQ setup: wq[%d]-id=%d, "
 				"parent cq[%d]-id=%d\n",
diff --git a/drivers/scsi/lpfc/lpfc_sli.c b/drivers/scsi/lpfc/lpfc_sli.c
index d7afd0fb457..982bd40353e 100644
--- a/drivers/scsi/lpfc/lpfc_sli.c
+++ b/drivers/scsi/lpfc/lpfc_sli.c
@@ -7796,14 +7796,14 @@ lpfc_sli4_bpl2sgl(struct lpfc_hba *phba, struct lpfc_iocbq *piocbq,
  *
  * Return: index into SLI4 fast-path FCP queue index.
  **/
-static uint32_t
+static inline uint32_t
 lpfc_sli4_scmd_to_wqidx_distr(struct lpfc_hba *phba)
 {
-	++phba->fcp_qidx;
-	if (phba->fcp_qidx >= phba->cfg_fcp_wq_count)
-		phba->fcp_qidx = 0;
+	int i;
 
-	return phba->fcp_qidx;
+	i = atomic_add_return(1, &phba->fcp_qidx);
+	i = (i % phba->cfg_fcp_wq_count);
+	return i;
 }
 
 /**
@@ -8323,16 +8323,6 @@ __lpfc_sli_issue_iocb_s4(struct lpfc_hba *phba, uint32_t ring_number,
 
 	if ((piocb->iocb_flag & LPFC_IO_FCP) ||
 		(piocb->iocb_flag & LPFC_USE_FCPWQIDX)) {
-		/*
-		 * For FCP command IOCB, get a new WQ index to distribute
-		 * WQE across the WQsr. On the other hand, for abort IOCB,
-		 * it carries the same WQ index to the original command
-		 * IOCB.
-		 */
-		if (piocb->iocb_flag & LPFC_IO_FCP)
-			piocb->fcp_wqidx = lpfc_sli4_scmd_to_wqidx_distr(phba);
-		if (unlikely(!phba->sli4_hba.fcp_wq))
-			return IOCB_ERROR;
 		if (lpfc_sli4_wq_put(phba->sli4_hba.fcp_wq[piocb->fcp_wqidx],
 				     &wqe))
 			return IOCB_ERROR;
@@ -8413,11 +8403,18 @@ int
 lpfc_sli_issue_iocb(struct lpfc_hba *phba, uint32_t ring_number,
 		    struct lpfc_iocbq *piocb, uint32_t flag)
 {
-	struct lpfc_sli_ring *pring = &phba->sli.ring[LPFC_ELS_RING];
+	struct lpfc_sli_ring *pring;
 	unsigned long iflags;
-	int rc;
+	int rc, idx;
 
 	if (phba->sli_rev == LPFC_SLI_REV4) {
+		if (piocb->iocb_flag &  LPFC_IO_FCP) {
+			if (unlikely(!phba->sli4_hba.fcp_wq))
+				return IOCB_ERROR;
+			idx = lpfc_sli4_scmd_to_wqidx_distr(phba);
+			piocb->fcp_wqidx = idx;
+			ring_number = MAX_SLI3_CONFIGURED_RINGS + idx;
+		}
 		pring = &phba->sli.ring[ring_number];
 		spin_lock_irqsave(&pring->ring_lock, iflags);
 		rc = __lpfc_sli_issue_iocb(phba, ring_number, piocb, flag);
@@ -8712,7 +8709,9 @@ lpfc_sli_setup(struct lpfc_hba *phba)
 	struct lpfc_sli *psli = &phba->sli;
 	struct lpfc_sli_ring *pring;
 
-	psli->num_rings = MAX_CONFIGURED_RINGS;
+	psli->num_rings = MAX_SLI3_CONFIGURED_RINGS;
+	if (phba->sli_rev == LPFC_SLI_REV4)
+		psli->num_rings += phba->cfg_fcp_eq_count;
 	psli->sli_flag = 0;
 	psli->fcp_ring = LPFC_FCP_RING;
 	psli->next_ring = LPFC_FCP_NEXT_RING;
@@ -11191,6 +11190,7 @@ lpfc_sli4_sp_handle_mcqe(struct lpfc_hba *phba, struct lpfc_cqe *cqe)
 /**
  * lpfc_sli4_sp_handle_els_wcqe - Handle els work-queue completion event
  * @phba: Pointer to HBA context object.
+ * @cq: Pointer to associated CQ
  * @wcqe: Pointer to work-queue completion queue entry.
  *
  * This routine handles an ELS work-queue completion event.
@@ -11198,12 +11198,12 @@ lpfc_sli4_sp_handle_mcqe(struct lpfc_hba *phba, struct lpfc_cqe *cqe)
  * Return: true if work posted to worker thread, otherwise false.
  **/
 static bool
-lpfc_sli4_sp_handle_els_wcqe(struct lpfc_hba *phba,
+lpfc_sli4_sp_handle_els_wcqe(struct lpfc_hba *phba, struct lpfc_queue *cq,
 			     struct lpfc_wcqe_complete *wcqe)
 {
 	struct lpfc_iocbq *irspiocbq;
 	unsigned long iflags;
-	struct lpfc_sli_ring *pring = &phba->sli.ring[LPFC_ELS_RING];
+	struct lpfc_sli_ring *pring = cq->pring;
 
 	/* Get an irspiocbq for later ELS response processing use */
 	irspiocbq = lpfc_sli_get_iocbq(phba);
@@ -11408,7 +11408,7 @@ lpfc_sli4_sp_handle_cqe(struct lpfc_hba *phba, struct lpfc_queue *cq,
 	case CQE_CODE_COMPL_WQE:
 		/* Process the WQ/RQ complete event */
 		phba->last_completion_time = jiffies;
-		workposted = lpfc_sli4_sp_handle_els_wcqe(phba,
+		workposted = lpfc_sli4_sp_handle_els_wcqe(phba, cq,
 				(struct lpfc_wcqe_complete *)&cqevt);
 		break;
 	case CQE_CODE_RELEASE_WQE:
@@ -11540,16 +11540,18 @@ lpfc_sli4_sp_handle_eqe(struct lpfc_hba *phba, struct lpfc_eqe *eqe)
 
 /**
  * lpfc_sli4_fp_handle_fcp_wcqe - Process fast-path work queue completion entry
- * @eqe: Pointer to fast-path completion queue entry.
+ * @phba: Pointer to HBA context object.
+ * @cq: Pointer to associated CQ
+ * @wcqe: Pointer to work-queue completion queue entry.
  *
  * This routine process a fast-path work queue completion entry from fast-path
  * event queue for FCP command response completion.
  **/
 static void
-lpfc_sli4_fp_handle_fcp_wcqe(struct lpfc_hba *phba,
+lpfc_sli4_fp_handle_fcp_wcqe(struct lpfc_hba *phba, struct lpfc_queue *cq,
 			     struct lpfc_wcqe_complete *wcqe)
 {
-	struct lpfc_sli_ring *pring = &phba->sli.ring[LPFC_FCP_RING];
+	struct lpfc_sli_ring *pring = cq->pring;
 	struct lpfc_iocbq *cmdiocbq;
 	struct lpfc_iocbq irspiocbq;
 	unsigned long iflags;
@@ -11667,7 +11669,7 @@ lpfc_sli4_fp_handle_wcqe(struct lpfc_hba *phba, struct lpfc_queue *cq,
 		cq->CQ_wq++;
 		/* Process the WQ complete event */
 		phba->last_completion_time = jiffies;
-		lpfc_sli4_fp_handle_fcp_wcqe(phba,
+		lpfc_sli4_fp_handle_fcp_wcqe(phba, cq,
 				(struct lpfc_wcqe_complete *)&wcqe);
 		break;
 	case CQE_CODE_RELEASE_WQE:
diff --git a/drivers/scsi/lpfc/lpfc_sli.h b/drivers/scsi/lpfc/lpfc_sli.h
index 2d64a2bab36..4b9b44e8e06 100644
--- a/drivers/scsi/lpfc/lpfc_sli.h
+++ b/drivers/scsi/lpfc/lpfc_sli.h
@@ -131,7 +131,9 @@ typedef struct lpfcMboxq {
 
 #define LPFC_MAX_RING_MASK  5	/* max num of rctl/type masks allowed per
 				   ring */
-#define LPFC_MAX_RING       4	/* max num of SLI rings used by driver */
+#define LPFC_SLI3_MAX_RING  4	/* Max num of SLI3 rings used by driver.
+				   For SLI4, an additional ring for each
+				   FCP WQ will be allocated.  */
 
 struct lpfc_sli_ring;
 
@@ -172,7 +174,7 @@ struct lpfc_sli3_ring {
 };
 
 struct lpfc_sli4_ring {
-	void *wqp;	/* Pointer to associated WQ */
+	struct lpfc_queue *wqp;	/* Pointer to associated WQ */
 };
 
 
@@ -284,7 +286,7 @@ struct lpfc_sli {
 #define LPFC_MENLO_MAINT          0x1000 /* need for menl fw download */
 #define LPFC_SLI_ASYNC_MBX_BLK    0x2000 /* Async mailbox is blocked */
 
-	struct lpfc_sli_ring ring[LPFC_MAX_RING];
+	struct lpfc_sli_ring *ring;
 	int fcp_ring;		/* ring used for FCP initiator commands */
 	int next_ring;
 
diff --git a/drivers/scsi/lpfc/lpfc_sli4.h b/drivers/scsi/lpfc/lpfc_sli4.h
index f4b57654787..e7d84134beb 100644
--- a/drivers/scsi/lpfc/lpfc_sli4.h
+++ b/drivers/scsi/lpfc/lpfc_sli4.h
@@ -142,6 +142,8 @@ struct lpfc_queue {
 	uint32_t host_index;	/* The host's index for putting or getting */
 	uint32_t hba_index;	/* The last known hba index for get or put */
 
+	struct lpfc_sli_ring *pring; /* ptr to io ring associated with q */
+
 	/* For q stats */
 	uint32_t q_cnt_1;
 	uint32_t q_cnt_2;